I have a nominal variable that divides my data in 8 groups, 1 to 8.

Another nominal variable divides my data in 2 groups, A and B.

Each combination of these two exists, 1A , 1B, 2A, 2B, etc.

The data are frequencies.

I assess whether group A is equal for each of the groups 1 to 8 via pairwise comparison using Fischer's exact test. (It is appropriate to assess A separately from B for the groups 1 to 8 and also Fischer's exact test is appropriate.)

I repeat the procedure with group B. I also do pairwise comparisons of groups 1 to 8 with the pooled data of groups A and B (this is also appropriate). All of this is appropriate for understanding what goes on in my survey.

Problem: Bonferroni correction for testing group A across pairwise comparisons of groups 1 to 8 would be 0.05/28 as 28 possible combinations exist. Fine so far. But I use a subset of my data, i.e., group A. With what number do I divide 0.05 with really when I assess group A, group B, and the pooled data of A and B?

Thank you in advance.